Gemini 3.1 Pro vs Grok 4.5
Gemini 3.1 Pro 
Grok 4.5 Verdict
- Gemini 3.1 Pro and Grok 4.5 are level on quality (63.9 vs 63.4).
- Gemini 3.1 Pro is stronger in agents & tools, composite, instruction following, long context, maths, multimodal, reasoning.
- Grok 4.5 is stronger in coding, human preference, knowledge.
- Grok 4.5 is 1.5× cheaper ($3.00 vs $4.50 per 1M blended).
- Gemini 3.1 Pro streams 2.0× faster (110 vs 56 tokens per second).
| Metric | Gemini 3.1 Pro | Grok 4.5 |
|---|---|---|
| BenchLeader Index | 63.9 | 63.4 |
| Agents & tools score | 63.7 | 58.6 |
| Coding score | 59.8 | 59.9 |
| Composite score | 67.2 | 66.0 |
| Human preference score | 59.6 | 66.9 |
| Instruction following score | 69.5 | – |
| Knowledge score | 66.1 | 76.0 |
| Long context score | 67.6 | 66.2 |
| Maths score | 60.7 | – |
| Multimodal score | 66.1 | 64.8 |
| Reasoning score | 69.5 | 68.9 |
| Blended price $/M | $4.50 | $3.00 |
| Output speed | 110 tok/s | 56 tok/s |
| Time to first answer | 30.5 s | 12.7 s |
| Context window | 1.0M | 500k |
| GPQA Diamond | 94.1% | – |
| FrontierMath Tiers 1–3 | 59.6% | – |
| FrontierMath Tier 4 | 26.8% | – |
| OTIS Mock AIME | 95.6% | – |
| Humanity's Last Exam | 46.4% | – |
| Terminal-Bench | 80.2% | – |
| SimpleBench | 79.6% | 70.0% |
| SciCode | 58.9% | – |
| WeirdML | 72.1% | 46.4% |
| APEX-Agents | 33.5% | 34.2% |
| FrontierCode | – | 42.4% |
| ProofBench | 26.0% | – |
| GSO-Bench | 22.6% | – |
| Epoch Capabilities Index | 155 | 153.9 |
| LMArena Text | 1487 | 1471 |
| LMArena Hard Prompts | 1507 | 1495 |
| LMArena Coding | 1521 | 1523 |
| LMArena WebDev | 1447 | 1556 |
| LMArena Vision | 1295 | 1291 |
| LMArena Agent | -5.3 | 3.9 |
| LiveBench | – | 75.8% |
| LiveBench Reasoning | – | 87.2% |
| LiveBench Coding | – | 68.6% |
| LiveBench Agentic Coding | – | 56.5% |
| LiveBench Mathematics | – | 90.8% |
| LiveBench Data Analysis | – | 73.0% |
| LiveBench Language | – | 82.8% |
| AA Intelligence Index | 30.4 | 39.1 |
| IFBench | 77.1% | – |
| AA-LCR | 82.0% | 79.3% |
| MMMU-Pro | 82.4% | 80.4% |
| AA-Omniscience | 31.9 | 25.3 |
| Terminal-Bench Hard | 53.8% | – |
| GPQA Diamond (AA) | 94.1% | 93.1% |
| Humanity's Last Exam (AA) | 47.0% | 42.7% |
| SciCode (AA) | 58.7% | 55.0% |
| τ²-Bench Telecom (AA) | 95.6% | – |
| AIME 2026 | 98.3% | – |
| HMMT February 2026 | 94.7% | – |
| MathArena Apex | 60.9% | – |
| MultiChallenge | 71.4% | – |
| PRBench Finance | 41.9% | – |
| PRBench Legal | 44.0% | – |
| MultiNRC | 64.7% | – |
| HiL-Bench | 35.3% | – |
| TutorBench | 53.0% | – |
| EQ-Bench 4 | 1142 | – |
| Kagi LLM Benchmark | – | 83.5% |
| ARC-AGI-1 | 98.0% | – |
| ARC-AGI-2 | 77.1% | – |
| ARC-AGI-3 | 0.4% | – |
Data as of 2026-09-09. Best configuration of each model; every score links to its source on the model pages.